Purpose: Support vector machine (SVM) model learning is commonly used in tumor prognosis evaluation for prostate cancer, breast cancer, glioma, etc. However, the influence of kernel paradigms, attribute numbers and data filtering used for SVM learning on IDH1 negative or positive prediction of glioma has not been investigated. Thus, here we focused on this issue and tried to optimize the automated tumor classifying system for IDH1 based on the histogram and texture features of pixel-level data derived from dynamic contrast enhancement MRI (DCE-MRI), diffusion weight image MRI (DWI-MRI) and 3-D arterial spin labeling (3D-ASL) images of glioma patients. Methods: 120 patients with a subsequent histologically confirmed diagnosis of glioma after surgery were included retrospectively. The individual parameters from DCE-MRI, DWI-MRI and 3D-ASL were calculated at the pixel level so as to extract the relevant histogram and texture features of the tumor. An SVM learning technique using leave one out cross-validation (LOOCV) was applied to this multimodal MR imaging-based histogram and texture features for the IDH1 classification. The kernels, dataset threshold, number of features for SVM model were manipulated to achieve a satisfying accuracy. Results: Using 1794 combined histogram and texture features from multimodal MRI parameters, the highest accuracy reached up to 84.3% for linear kernel (c=0.125) and 94.3% for RBF kernel (gamma=1/32 and c=2) SVM, respectively. Based on the SVM RFE ranking result, the satisfying accuracy was obtained when about 250~500 mostly weighted features were selected. Besides, when the data filtering strategy was applied, the accuracy of linear SVM increased up to 87.5% with about 35% 1-side or 25% 2-side data cutoff. Conclusion: The influence of kernel paradigms, attribute numbers and data filtering on the accuracy of SVM learning for IDH1 prediction cannot be ignored. It should be considered in optimizing IDH1 prediction model for future clinical use.